Features
- Long text sent to the masking model is now processed in overlapping pieces, one piece at a time, so longer content can be handled without being cut off.
Bug Fixes
- Masking time no longer counts toward the 2-minute routing window.
Improvements
- A GPU that is too slow for a request is now retried, and a disconnected GPU can recover without restarting the service.
- The masking service processes AI model work in 2048-token chunks by default, with an option to change the chunk size if needed.
Install
Download from usemagpie.ai, or curl -fsSL https://usemagpie.ai/install.sh | sh. An installed magpie updates itself.
- macOS (Apple Silicon):
magpie-darwin-arm64.dmg - macOS (Intel):
magpie-darwin-amd64.dmg - Windows:
magpie-windows-amd64.exe(ARM:magpie-windows-arm64.exe) - Linux:
magpie-linux-amd64/magpie-linux-arm64(needs GTK 3 and WebKitGTK 4.1) - Terminal only:
magpie-cli-<os>-<arch>
功能
- 发送到掩码模型的长文本现在会以重叠分块的方式逐块处理，因此可以处理更长的内容而不会被截断。
错误修复
- 掩码处理时间不再计入 2 分钟路由窗口。
改进
- 对于某个请求而言速度过慢的 GPU 现在会被重试，并且断开连接的 GPU 可以在不重启服务的情况下恢复。
- 掩码服务默认以 2048 个 token 的分块处理 AI 模型工作，并可根据需要更改分块大小。
安装
从 usemagpie.ai 下载，或运行 curl -fsSL https://usemagpie.ai/install.sh | sh。已安装的 magpie 会自动更新。